[llvm] [AMDGPU][InsertWaitCnts][NFC] Merge VMEM_ACCESS and VMEM_READ_ACCESS into a single event type (PR #171973)

[Pankaj Dwivedi via llvm-commits]

https://github.com/PankajDwivedi-25 created https://github.com/llvm/llvm-project/pull/171973

This patch merges VMEM_READ_ACCESS  and VMEM_ACCESS into a single VMEM_READ_ACCESS event type, simplifying
the code without changing functionality.

>From 3660338fe4c8a8cd064f0df9623086895e9a1ff4 Mon Sep 17 00:00:00 2001
From: Pankaj kumar divedi <Pankajkumar.divedi at amd.com>
Date: Fri, 12 Dec 2025 13:44:01 +0530
Subject: [PATCH] merge v_read_access and v_mem_access because no subtarget use
 both

---
 llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
index 70db7b4918515..53ad7d81e3d08 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -111,7 +111,6 @@ struct HardwareLimits {
 };
 
 #define AMDGPU_DECLARE_WAIT_EVENTS(DECL)                                       \
-  DECL(VMEM_ACCESS)              /* vmem read & write */                       \
   DECL(VMEM_READ_ACCESS)         /* vmem read */                               \
   DECL(VMEM_SAMPLER_READ_ACCESS) /* vmem SAMPLER read (gfx12+ only) */         \
   DECL(VMEM_BVH_READ_ACCESS)     /* vmem BVH read (gfx12+ only) */             \
@@ -362,8 +361,8 @@ class WaitcntGeneratorPreGFX12 : public WaitcntGenerator {
     assert(ST);
 
     static const unsigned WaitEventMaskForInstPreGFX12[NUM_INST_CNTS] = {
-        eventMask({VMEM_ACCESS, VMEM_READ_ACCESS, VMEM_SAMPLER_READ_ACCESS,
-                   VMEM_BVH_READ_ACCESS}),
+        eventMask(
+            {VMEM_READ_ACCESS, VMEM_SAMPLER_READ_ACCESS, VMEM_BVH_READ_ACCESS}),
         eventMask({SMEM_ACCESS, LDS_ACCESS, GDS_ACCESS, SQ_MESSAGE}),
         eventMask({EXP_GPR_LOCK, GDS_GPR_LOCK, VMW_GPR_LOCK, EXP_PARAM_ACCESS,
                    EXP_POS_ACCESS, EXP_LDS_ACCESS}),
@@ -399,7 +398,7 @@ class WaitcntGeneratorGFX12Plus : public WaitcntGenerator {
     assert(ST);
 
     static const unsigned WaitEventMaskForInstGFX12Plus[NUM_INST_CNTS] = {
-        eventMask({VMEM_ACCESS, VMEM_READ_ACCESS}),
+        eventMask({VMEM_READ_ACCESS}),
         eventMask({LDS_ACCESS, GDS_ACCESS}),
         eventMask({EXP_GPR_LOCK, GDS_GPR_LOCK, VMW_GPR_LOCK, EXP_PARAM_ACCESS,
                    EXP_POS_ACCESS, EXP_LDS_ACCESS}),
@@ -549,7 +548,7 @@ class SIInsertWaitcnts {
     // LDS DMA loads are also stores, but on the LDS side. On the VMEM side
     // these should use VM_CNT.
     if (!ST->hasVscnt() || SIInstrInfo::mayWriteLDSThroughDMA(Inst))
-      return VMEM_ACCESS;
+      return VMEM_READ_ACCESS;
     if (Inst.mayStore() &&
         (!Inst.mayLoad() || SIInstrInfo::isAtomicNoRet(Inst))) {
       // FLAT and SCRATCH instructions may access scratch. Other VMEM